MODELLING CAUSAL MECHANISMS IN ORGANISMAL AGING

Abstract

Geroscience seeks to clarify and explain the connection between chronological age and declining organism health. The field is rich in complex phenomena and increasingly rich in large, multivariate datasets as well, making it fertile ground for quantitative modelling. This thesis outlines the steps needed to develop such models, from contending with aging study data to validating a causal model of effective dynamics. Ultimately, the goal of this thesis is make interpretable predictions for the causes and effects of organism aging, with special attention paid to humans. This goal is met by developing and applying the Stochastic Finite (SF) difference model which uses an interaction network to predict multivariate, future values based on current values. The process by which aging study data facilitates this type of model building involves multiple pitfalls and key assumptions which are addressed in introductory chapters on missing data and unsupervised learning of aging metrics. The SF model is then validated on two mouse datasets and two human datasets, unveiling characteristic dynamical behaviour of aging systems. In particular, we observe mallostasis: the steady-state drift of biomarker values towards worse health. The approach is then applied to specific age metrics, so called “biological ages”, to probe more directly the sequence of events which occurs during natural aging. This provides insight into the strengths and weaknesses of existing qualitative theories and provides a path towards more quantitative theories of aging.